Semantic Class Distribution Learning for Debiasing Semi-Supervised Medical Image Segmentation

TL;DR

This paper introduces the SCDL framework to improve medical image segmentation by learning structured class-conditional features, effectively addressing class imbalance and enhancing minority class segmentation.

Contribution

The paper proposes a novel plug-and-play module, SCDL, that aligns class embeddings and uses semantic anchors to mitigate bias in semi-supervised medical image segmentation.

Findings

SCDL achieves state-of-the-art results on Synapse and AMOS datasets.

Significant improvements in minority class segmentation performance.

Effective in both overall and class-level metrics.

Abstract

Medical image segmentation is critical for computer-aided diagnosis. However, dense pixel-level annotation is time-consuming and expensive, and medical datasets often exhibit severe class imbalance. Such imbalance causes minority structures to be overwhelmed by dominant classes in feature representations, hindering the learning of discriminative features and making reliable segmentation particularly challenging. To address this, we propose the Semantic Class Distribution Learning (SCDL) framework, a plug-and-play module that mitigates supervision and representation biases by learning structured class-conditional feature distributions. SCDL integrates Class Distribution Bidirectional Alignment (CDBA) to align embeddings with learnable class proxies and leverages Semantic Anchor Constraints (SAC) to guide proxies using labeled data. Experiments on the Synapse and AMOS datasets demonstrate that SCDL significantly improves segmentation performance across both overall and class-level metrics, with particularly strong gains on minority classes, achieving state-of-the-art results. Our code is released at https://github.com/Zyh55555/SCDL.